Fast and Scalable Input/Output

Format independent solution

Input

Parallel loading of unstructured mesh databases

Our I/O module is designed to efficiently load sequential mesh database files and perform high-quality domain decomposition. It is developed from ground up as a highly parallel tool designed for HPC environment including the high-performance storage systems. It is composed of several commonly used approaches that together lead to a robust solution.​

The user can use the same database file stored in his favorite format for multiple solutions. Each solution can be executed on a different number of compute nodes without any additional penalty because our approach does not require to run every time with a fixed number of MPI process, and domains, as in case of a parallel binary file. This allows the engineer to use a different amount of computational resources based on his needs or availability.

  • Supported formats
    • OpenFOAM
    • NetGen
    • EnSight
    • VTK
    • ANSYS_CDB
    • Abaqus
  • New format support can be developed based on customer requests.

ESPRESO parallel input workflow scalability

mesh processing - data distribution - domain decomposition

Jet Engine

sequential ASCII FEM database 150 GB
822 milion mesh nodes
mesh loaded and decomposed in 14 seconds

384 of CPU Cores
208 seconds
768 of CPU Cores
105 seconds
1,536 of CPU Cores
56 seconds
3,072 of CPU Cores
36 seconds
6,144 of CPU Cores
28 seconds
12,288 of CPU Cores
14

Workflow

PreProcessing

Create unstructured meshes in your favorite robust and efficient preProcessing tool as you are used to and Export database file from preProcessing tool in your preferable format

bonn

ESPRESO Scalable I/O

Fast parallel processing and hybrid domain decomposition of unstructured meshes with direct connection to the massively parallel solvers

Datový zdroj 3

Standard Solution

In common open Source approaches, the sequential converting to parallel native format is needed. This create sequential barrier for scalable solvers.

Supercomputer

Solve the problems at the top of your database files

Asynchronous Output

For the PostProcessing purpose, storing result outputs to commonly used PostProcessing formats are implemented.

Output to commonly used postProcessing formats EnSight and VTK

Overlapping ongoing computation by storing solution results

Result monitoring of selected regions for statistic and optimization toolchains

Solution restarting without ties to the previous number of resources